RE: size too large for range of Octave's index type

[Tim Rueth]

> -----Original Message-----
> From: Jaroslav Hajek [mailto:address@hidden 
> Sent: Thursday, December 02, 2010 12:43 PM
> To: address@hidden
> Cc: address@hidden
> Subject: Re: size too large for range of Octave's index type
> 
> On Thu, Dec 2, 2010 at 7:44 AM, Tim Rueth <address@hidden> wrote:
> > I hit this problem a while back, and fixed it by changing my 
> > double-precision vars to single precision.  But now in my 
> current sim, 
> > my ndgrid matrices are so large, even with single 
> precision, that I'm 
> > getting the following error again:
> >
> > error:  memory exhausted or requested size too large for range of 
> > Octave's index type
> >
> > After getting this error, I did a "whos" and got:
> >
> > 421605065 elements using 1515209069 bytes
> >
> > Okay, that's 1.41 GB.  What is Octave's limit?  In reading previous 
> > posts, it looks like Octave's limit is 2e9 elements or 2GB, but I'm 
> > not sure what build this refers to, in case later builds 
> support more.  
> > I'm running version 3.2.3 on Win7 w/6GB of RAM.  I also tried it on 
> > 3.2.4 and got the same error.
> >
> 
> In 3.2.3, the limit is 2GB memory. In 3.3.5x+, this is 
> already fixed to 2e9 elements (unless compiled with --enable-64).
> Note that this is a limit for a single array (that's the 
> "requested size too large" part) - you can have 10 1GB arrays 
> if your OS allows you to.
> 

Hm-mm.  But the memory requirement for all of my vars is only 1.41GB (< 2GB)
with fewer than 2e9 elements total, but I'm still getting the error.  So I'm
still confused why I'm getting the error.

--Tim